{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TULFDY5NNTRN6MMYSQLHZQPX3V","short_pith_number":"pith:TULFDY5N","schema_version":"1.0","canonical_sha256":"9d1651e3ad6ce2df319894167cc1f7dd4d366ebe09b10ca359cddcccf8d859d3","source":{"kind":"arxiv","id":"2506.18879","version":1},"attestation_state":"computed","paper":{"title":"CommVQ: Commutative Vector Quantization for KV Cache Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chong Wang, Chuang Gan, Colorado Reed, Foroozan Karimzadeh, Junyan Li, Muhammad Yusuf Hassan, Pengsheng Guo, Talha Chafekar, Tianle Cai, Yang Zhang, Zhile Ren","submitted_at":"2025-06-23T17:50:11Z","abstract_excerpt":"Large Language Models (LLMs) are increasingly used in applications requiring long context lengths, but the key-value (KV) cache often becomes a memory bottleneck on GPUs as context grows. To address this, we propose Commutative Vector Quantization (CommVQ) to significantly reduce memory usage for long-context LLM inference. We first introduce additive quantization with a lightweight encoder and codebook to compress the KV cache, which can be decoded via simple matrix multiplication. To further reduce computational costs during decoding, we design the codebook to be commutative with Rotary Posi"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2506.18879","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-23T17:50:11Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"66e984959dc596d00fc3068e6d706246fbd1b2fe95d3d54b31e008ff798194ca","abstract_canon_sha256":"450cd49672b36ebc35074691d8e863ff8384c19eb131a951c679d07c79bb3fbf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:25:57.425520Z","signature_b64":"kT4CrhtYO/2eudxVGhL/n2nqbpLWlBwS75NLdbDqeArz48FciF7PYZ0NSETIZowP3EtA4ZA18sd8Vdee+vr6CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9d1651e3ad6ce2df319894167cc1f7dd4d366ebe09b10ca359cddcccf8d859d3","last_reissued_at":"2026-07-05T11:25:57.425036Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:25:57.425036Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CommVQ: Commutative Vector Quantization for KV Cache Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chong Wang, Chuang Gan, Colorado Reed, Foroozan Karimzadeh, Junyan Li, Muhammad Yusuf Hassan, Pengsheng Guo, Talha Chafekar, Tianle Cai, Yang Zhang, Zhile Ren","submitted_at":"2025-06-23T17:50:11Z","abstract_excerpt":"Large Language Models (LLMs) are increasingly used in applications requiring long context lengths, but the key-value (KV) cache often becomes a memory bottleneck on GPUs as context grows. To address this, we propose Commutative Vector Quantization (CommVQ) to significantly reduce memory usage for long-context LLM inference. We first introduce additive quantization with a lightweight encoder and codebook to compress the KV cache, which can be decoded via simple matrix multiplication. To further reduce computational costs during decoding, we design the codebook to be commutative with Rotary Posi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.18879","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2506.18879/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2506.18879","created_at":"2026-07-05T11:25:57.425093+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.18879v1","created_at":"2026-07-05T11:25:57.425093+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.18879","created_at":"2026-07-05T11:25:57.425093+00:00"},{"alias_kind":"pith_short_12","alias_value":"TULFDY5NNTRN","created_at":"2026-07-05T11:25:57.425093+00:00"},{"alias_kind":"pith_short_16","alias_value":"TULFDY5NNTRN6MMY","created_at":"2026-07-05T11:25:57.425093+00:00"},{"alias_kind":"pith_short_8","alias_value":"TULFDY5N","created_at":"2026-07-05T11:25:57.425093+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.08382","citing_title":"STAR-KV: Low-Rank KV Cache Compression via Soft Thresholding for Adaptive Rank Control","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24144","citing_title":"EVA: Accelerating LLM Decoding via an Efficient Vector Quantization Architecture","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2602.03922","citing_title":"Online Vector Quantized Attention","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2602.02958","citing_title":"Quant VideoGen: Auto-Regressive Long Video Generation via 2-Bit KV-Cache Quantization","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19157","citing_title":"SAW-INT4: System-Aware 4-Bit KV-Cache Quantization for Real-World LLM Serving","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TULFDY5NNTRN6MMYSQLHZQPX3V","json":"https://pith.science/pith/TULFDY5NNTRN6MMYSQLHZQPX3V.json","graph_json":"https://pith.science/api/pith-number/TULFDY5NNTRN6MMYSQLHZQPX3V/graph.json","events_json":"https://pith.science/api/pith-number/TULFDY5NNTRN6MMYSQLHZQPX3V/events.json","paper":"https://pith.science/paper/TULFDY5N"},"agent_actions":{"view_html":"https://pith.science/pith/TULFDY5NNTRN6MMYSQLHZQPX3V","download_json":"https://pith.science/pith/TULFDY5NNTRN6MMYSQLHZQPX3V.json","view_paper":"https://pith.science/paper/TULFDY5N","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.18879&json=true","fetch_graph":"https://pith.science/api/pith-number/TULFDY5NNTRN6MMYSQLHZQPX3V/graph.json","fetch_events":"https://pith.science/api/pith-number/TULFDY5NNTRN6MMYSQLHZQPX3V/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TULFDY5NNTRN6MMYSQLHZQPX3V/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TULFDY5NNTRN6MMYSQLHZQPX3V/action/storage_attestation","attest_author":"https://pith.science/pith/TULFDY5NNTRN6MMYSQLHZQPX3V/action/author_attestation","sign_citation":"https://pith.science/pith/TULFDY5NNTRN6MMYSQLHZQPX3V/action/citation_signature","submit_replication":"https://pith.science/pith/TULFDY5NNTRN6MMYSQLHZQPX3V/action/replication_record"}},"created_at":"2026-07-05T11:25:57.425093+00:00","updated_at":"2026-07-05T11:25:57.425093+00:00"}